How is data isolated between customers?
Ask: at which layer is tenant isolation enforced; do background jobs run with the same scope as user requests; is the AI layer, meaning memory, drafts, embeddings, and logs, isolated the same way; are properties within an organization separated; who inside the vendor can cross the boundary and under what controls. A good answer says data layer, every query, including jobs, and names the property boundary.
ftes.ai: every record is scoped to an organization and every query that reaches the database carries it, including background jobs; properties within an organization are separate with their own memory, targets, and setup; memory, drafts, and credentials are scoped the same way.
- 1. Isolation layer? 2. Jobs scoped? 3. AI layer scoped? 4. Property separation? 5. Internal access controls?
How are the credentials we grant handled?
Ask: how are integration credentials and API keys stored; do they ever appear in responses, logs, or the interface; how are they scoped; how are they revoked and how fast; how does the vendor's own app authenticate. A good answer says encrypted at rest with a separately held key, never exposed, scoped per property, revocable from one place, logged.
ftes.ai: credentials are encrypted before storage with a key held separately from the data, never appear in responses or logs, are granted per organization and per property, and any integration can be disconnected from the workspace at any time; the app uses short-lived access tokens with a rotating httpOnly refresh cookie, and the browser never talks to the API directly.
- 6. Storage? 7. Exposure? 8. Scope? 9. Revocation? 10. App auth?
Which model providers process our data, and on what terms?
Ask: which providers process the data; does the vendor train models on customer data; can we bring our own provider key; are the provider's data terms passed through in writing; where is data processed and stored. A good answer names the provider, says no to training, offers BYOK, and points to a data processing agreement listing sub-processors.
ftes.ai: work runs through the Claude Agent SDK; the platform does not train models on customer data; organizations can bring their own provider key; its GDPR page states its position on processing, transfers, and sub-processors, to be confirmed in writing rather than from any article.
- 11. Providers? 12. Training on our data? 13. BYOK? 14. Terms in writing? 15. Processing location?
Which governance mechanisms exist, and can we see them?
Ask: does every action start gated behind a named person; is autonomy granted per class of action from a record or by a single trust setting; what is the never-automated list, and can we see it; how quickly can autonomy be withdrawn; how are unsourced claims prevented from reaching a person. A good answer shows a workspace with an approval queue, a per-class ladder, a fixed list, and a fact gate.
ftes.ai: every Digital FTE starts at Level 0 with every action gated; autonomy is per class from the approval record and stepped back in one logged change; the never-automated list covers decisions about a person's outcome and their money-and-contract equivalents; the fact gate blocks unsourced claims before approval.
- 16. Gated by default? 17. Per-class or single setting? 18. Never-automated list? 19. Withdrawal speed? 20. Unsourced claims?
What does the audit trail contain, and can we export it?
Ask: what the trail records for each action; whether it covers management actions such as autonomy grants, budget changes, and credential changes; whether it can be exported; and whether spend is a ledger. A good answer lists proposal, evidence, decision, reason, approver, time, autonomy level, effect, and management changes, and says yes to export.
ftes.ai: every proposal, approval, publish, and token spent is logged with who, what, and when, with evidence and the autonomy level; management changes are logged; token spend is an immutable ledger; the trail can be exported.
- 21. Trail contents? 22. Management actions? 23. Export?
What happens at the end?
Ask: how integrations are disconnected, how data deletion is requested and confirmed, what is retained and for how long, and whether the organization can keep its audit trail and confirmed memory. A good answer gives a disconnect path in the product, a named deletion contact, retention terms in the agreement, and export of the record. Ask this before signing, not before leaving.
ftes.ai: any integration can be disconnected from the workspace at any time; deletion of an organization's data is requested through the privacy contact on its security page; the audit trail and confirmed Brand Brain can be retained or exported.
- 24. Disconnect, deletion, retention, and what we keep?
How do you use this questionnaire?
Send all twenty-four questions in writing to every vendor under consideration, including ftes.ai, with a deadline, and file the answers with the procurement record. Read each answer for a mechanism: a place in the product, a layer in the architecture, a document you can hold. Mark every adjective without a mechanism and every certification offered in place of an answer; those are findings, not disqualifications, but a vendor with many of them has not built what you are asking about.
Then verify the two or three answers you can verify yourself: fetch a page as a scriptless crawler, connect one tool and disconnect it, trace one action in the trail. A vendor whose written answers survive verification is one worth piloting.